search

eloialonso / iris

Transformers are Sample-Efficient World Models. ICLR 2023, notable top 5%.
https://openreview.net/forum?id=vhFu1Acb0xb
GNU General Public License v3.0
805 stars 80 forks source link

About the study restart function #17

Closed tatematsu123 closed 1 year ago

tatematsu123 commented 1 year ago

I would like to restart an interrupted training using the following code in src/trainer.py.

if cfg.common.resume:.
    self.load_checkpoint()

What directory and what command should I type to use this function?

I have changed cfg.common.resume to true in outputs/YYYYY-MM-DD/hh-mm-ss/ and I run python src/main.py env.train.id=BreakoutNoFrameskip-v4 common.device=cuda:0 wandb.mode=online, but I got an error.

eloialonso commented 1 year ago

You just need to run ./scripts/resume.sh in the run folder outputs/YYYY-MM-DD/hh-mm-ss/.

I hope that helps!

tatematsu123 commented 1 year ago

Thank you!